nVidia is turning their CUDA GPU architecture into a desktop computer of the supercomputing variety. ASUS, Dell, and Lenovo are apparently shipping boxes based on nVidia's Tesla C1060 GPU -- "the power of a traditional supercomputer cluster at 1/100th of the price."
So which mainstream CAD package will be the first? 'Course, it won't involve a simple re-compile, like it does with well-written code being ported to another operating system; this is a non-Intel-based architecture.
Or is this just a pr stunt by nVidia?
I would expect rendering and FEA to run on something like this far before a CAD package. nVidia's Tesla or AMD's FireStream are both massively parallel computing machines. They will only be advantageous for software that is massively parrallel. As far as I know - CAD is not even close yet. FEA and rendering are already written for that. I do know that starting with SolidWorks 2008, some PhysX components get installed. PhysX was bought buy nVidia and the software can be powered by regular gaming cards. The latest gaming cards have enormous parallel compute capabilities for much less cost than a firestream or tesla system. I think the PhysX engine is being used during the most simplistic of motion simulations in SolidWorks (friction and certain contact parameters are ingnored - which greatly affect solution time). Of course, FEA and Rendering cold also be done in the cloud at a much much lower cost to the user (they don't have to purchase the hardware at all)
.
Posted by: Pete Yodis | Nov 18, 2008 at 12:23 PM